caveman

Record a problem working with an agent as one entry in the current repo's agent work log.

Explain one specific problem as bullets under Expected behavior, Actual behavior (symptoms), Root cause, and Fix, in plain words.

Summarize what was done, what is planned, or what is still open as nested one-line bullets.

Teaches a topic in two beats: show and explain in chat, then the human writes a small version to internalize.

Behavioral guidelines to reduce common LLM coding mistakes.

Mines Cursor chat transcripts and git changes for short AI blog-post ideas.

Guide the human through a specific task one atomic step at a time, waiting for confirmation before advancing.

Create GitHub releases following the established version scheme, title format, and release-notes style.

Update the portfolio site and chat knowledge from the canonical Markdown CV.

Hand learning-relevant implementation work to the human (solo or guided coaching), then review and iterate.

Create or adapt a skill in this skills collection, including a mandatory malice/safety review of SKILL.md, scripts, and companion files.

Execute a written implementation plan from docs/plans/ one task at a time, hard-stopping after each task for a small reviewable PR.